AI Plugins And Skills

Expert SQL developer specializing in complex query optimization, database design, and performance tuning across PostgreSQL, MySQL, SQL Server, and Oracle. Masters advanced SQL features, indexing strategies, and data warehousing patterns.

Route a task to a Low/Medium/High spec tier by stakes, then enforce the matching spec depth, review gates, and calibration. Use when starting any plan or spec, deciding how much detail a plan needs, onboarding developers to the planning standard, or when unsure whether a spec is over- or under-specified. Invoke BEFORE plan-workflow so it sets the depth plan-workflow then produces.

Use when running ad-hoc read-only T-SQL against Azure SQL — exploring Query Store, sampling tables, inspecting DMVs, joining sys.* views, or producing JSON/CSV for an alerting pipeline. Read-only by construction (AST-validated, rolled-back transaction, no INSERT/UPDATE/DELETE/EXEC/OPENROWSET). Returns TOON by default for token-efficient consumption by Claude; --format=json|table|csv for other consumers.

Use when running ad-hoc read-only SQL against Azure Database for PostgreSQL — exploring tables, sampling rows, checking key uniqueness/overlap across tables, inspecting pg_stat_* / pg_catalog, or producing JSON/CSV for a pipeline. Read-only by construction (server-enforced read-only transaction + client-side guard, always rolled back; no INSERT/UPDATE/DELETE/DDL/COPY/dblink). Returns TOON by default for token-efficient consumption by Claude; --format=json|table|csv for other consumers. This is the PostgreSQL counterpart to sql-query (which targets Azure SQL only).

Python backend expert. PROACTIVELY use when working with Django, FastAPI, Flask, Pydantic, SQLAlchemy, or async Python APIs. Covers backend patterns, data validation, and ORM best practices.

Use when a refactor needs the same identifier rename applied across many files (type-cluster renames, magic-string vocabulary moves) — runs a single bash invocation that does longest-first, word-boundary safe replacements across all tracked files in scope, instead of issuing dozens of per-file Edit tool calls

Manages planning documentation through the kanban board at planning/. Use when creating, progressing, or completing plans to ensure every document lands in the right directory (planning/, in-progress/, completed/, not-implemented/, blocked/, backlog/, reminders/, waves/, evidence/, needs-review/) and stays in source control alongside the code it describes. Also covers parallel tracks — one in-progress plan per track when independent plans run in sub-agents. Pairs with plan-writing-syntax, which owns what is written inside the document; this owns where the document lives and when it moves.

Composes and executes a standardized, spec-driven, test-first .NET feature prompt. Invoke as /prompt-dotnet followed by the feature description. Chains spec-tier (stakes tiering + review gates), test-filter-development (RED-GREEN-REFACTOR with captured failure output and a seeded mutant), and csharp-quality-developer; if the harness lacks those skills, the inlined discipline is the contract. Oriented to ASP.NET Core services/APIs (the integration test uses WebApplicationFactory). Use when implementing a .NET feature and you want tiered spec depth, an evidenced test filter, and mandatory review.

Records a task a person must actively do, which no test or gate can catch. Use when a plan names work that needs a human, a machine, a conversation or a calendar — a check on hardware nobody on the team owns, a decision only one role can take, something to raise at the next session — and the plan alone would leave it to memory. Sits beside plan-workflow; it never schedules work, it remembers the work software cannot verify.

Angular 20+/TypeScript frontend expert. PROACTIVELY use when working with Angular components, signals, RxJS, NgRx, angular.json configuration, or ng serve issues.

Secrets hygiene for .NET development. Use when handling credentials, connection strings, API keys, or any secret value — especially when setting up local dev environments, configuring user-secrets, or bootstrapping Azure App Configuration + Key Vault access. Activates on any prompt mentioning secrets, credentials, connection strings, user-secrets, AZURE_CLIENT_SECRET, AZURE_CLIENT_ID, or AppConfig connection strings.

TypeScript best practices expert. PROACTIVELY use when working with TypeScript/JavaScript files (.ts, .tsx, .js, .jsx), type errors, ESLint issues, or strict mode configuration. Covers patterns, ESLint best practices, and strict type handling.

Generate a deterministic architecture map of Python, C#, and TypeScript codebases using AST parsing. Outputs a token-efficient structure section for CLAUDE.md. Use when mapping a codebase, generating architecture documentation, updating claude.md structure, or onboarding to a new project.

Advanced Context Engineering with Frequent Intentional Compaction (ACE-FCA) for complex coding tasks. Use this skill when working on brownfield codebases, large repos (100k+ LOC), complex bugs, multi-file refactors, or features requiring deep codebase understanding. Triggers include requests to fix bugs in unfamiliar code, implement features in large codebases, understand complex code flows, or when the user mentions "research plan implement", "ACE-FCA", or "frequent compaction". This workflow prevents context window exhaustion and produces high-quality, reviewable artifacts.

Use when queueing a slice of a branch for code review — builds a slice spec from a commit range, focus areas, and plan-doc refs, and drops it in planning/code-reviews/needs-review/ for the intake loop to pick up

Use when you have authored a plan and want the other AI agent (Claude or Copilot) to review it - copies the active plan to planning/needs-review/ for adversarial review

React best practices expert. PROACTIVELY use when working with React components, JSX, hooks (useState, useEffect), state management, or component architecture and performance optimization.

Expert Angular architect mastering Angular 20+ with enterprise patterns. Specializes in signals, RxJS, NgRx state management, micro-frontend architecture, and performance optimization with focus on building scalable enterprise applications.

Use when reviewed code-review slices are ready for triage and the developer wants to walk findings, classify them accept/merge/reject/flag, and apply accepted changes to code — third punch-in command of the code review workflow, after code-review-plan-execute completes

Use on a loop interval to check for code slices awaiting review in planning/code-reviews/needs-review/, move to in-progress, dispatch code-reviewer subagent, append findings, and move to reviewed/ - silent when no files found

Use on a loop interval to check for plans awaiting review in planning/needs-review/, move to in-progress, dispatch plan-reviewer subagent, append feedback, and move to reviewed/ - silent when no files found

Records what a plan review actually decided — accepted, merged, rejected, flagged — so the reasoning outlives the conversation. Use immediately after evaluating a review with plan-receive-review, before moving on. Closes the gap where a review is acted on and nothing in the repository says which findings were taken or why the others were not.

Turns planning/blocked/ into an agenda for a session that could actually clear it. Use when something is briefly available — a device, an environment, an hour with the one person who can decide, a maintenance window — and the question is "what do I do with it?". Groups by blocker rather than by plan, orders the work inside each group, cites where each step is already specified, and names what each one unblocks.

Use when a plan you authored has been reviewed and is waiting in planning/needs-review/reviewed/ - retrieves reviewed plan, evaluates each review point with technical rigor, and presents findings with recommended actions

Finds plans that are stuck, records what they are stuck on, moves them to planning/blocked/, and files a reminder for each blocker something outside the repository has to clear. Use when the in-progress column no longer answers "what is actually running" — when a plan has not moved in a while, when a status section contradicts its own frontmatter, or before scheduling, so that the schedule reads a board that is telling the truth.

Use when writing or editing any planning document — a plan, a backlog stub, a reminder, a review or a wave — to get its frontmatter right. Owns the shape of the document — the kind and status fields, the per-kind status vocabularies, what each status is required to carry, and cross-references written as names rather than paths. Pairs with plan-workflow, which owns which directory a document lives in and when it moves. Use it before writing the first line, not after a validator refuses the file.

Azure cloud infrastructure expert specializing in network design, identity integration, PowerShell automation with Az modules, and infrastructure-as-code patterns using Bicep.

Use when initializing a new Python project or adding quality tooling to an existing one - sets up ruff linter and formatter, pyright type checker, and pre-commit hooks with opinionated defaults that enforce python-quality-developer standards

Code Review Plan Create

Use when a code-review meta-plan is finalized and you need to fan it out into one slice spec per row in the meta-plan's slice table — produces N files in planning/code-reviews/needs-review/ ready for code-intake-review to drain

Code Synthesize Reviews

Use when implementing any feature or bugfix with AI assistance, before writing implementation code. RED → GREEN → REFACTOR with the filter made load-bearing - captured failure output rather than attestation, a seeded mutant after GREEN, and rejection when the suite survives it. Use in place of test-driven-development, which names the ordering rather than the mechanism that carries the value.

Use when a code-review meta-plan is ready and you want to drive the full review pipeline — pre-flight, seed slices, dispatch reviewers per slice, synthesize findings, and notify on completion; defensive and idempotent so repeated runs resume from where the last run stopped

Enforce C# coding standards including StyleCop rules (SA1028, SA1518, SA1101, etc.), file formatting (CRLF, trailing whitespace), naming conventions (this. prefix, no underscore prefix), XML documentation, LoggerMessage patterns, and brace placement. Use when writing C# code, reviewing .cs files, fixing StyleCop violations, or ensuring code quality compliance.

Enforce Python coding standards including ruff rules, pyright type safety, avoiding Any, consistent return types, proper HTTP error signaling, exception chaining, async-first patterns, typed dicts for structured data, and parameter ordering. Use when writing Python code, reviewing .py files, fixing linter violations, or ensuring code quality compliance.

Use when peeking, draining, snapshotting metrics on, enumerating, or otherwise inspecting/manipulating Azure Service Bus subscriptions, queues, or dead-letter sub-queues — including diagnosing DLQ accumulation, capturing parked messages for forensics before discard, draining stuck alert backlogs, comparing active/dlq counts across multiple subs, enumerating an entire namespace's topology, or clearing a subscription's ForwardTo (which the az CLI cannot do).